Lecture 11. Probability Theory: an Overveiw

Transcription

1 Math Mathematical Statistics Lecture 11. Probability Theory: an Overveiw February 11, 2013 Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

2 The starting point in developing the probability theory is the notion of a sample space = the set of possible outcomes. The sample space Ω is the set of possible outcomes of an experiment Points ω Ω are called realizations Events are subsets of Ω Next, to every event A Ω, we assign a real number P(A), called the probability of A. We call function P : {subsets of Ω} R a probability distribution. Function P is not arbitrary, it satisfies several natural properties (called axioms of probability): 1 0 P(A) 1 (Events range from never happening to always happening) 2 P(Ω) = 1 (Something must happen) 3 P( ) = 0 (Nothing never happens) 4 P(A) + P(Ā) = 1 (A must either happen or not-happen) 5 P(A + B) = P(A) + P(B) P(AB) Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

3 Statistical Independence Two events A and B are independent if P(AB) = P(A)P(B) Independence can arise in two distinct ways: 1 We explicitly assume that two events are independent. 2 We derive independence of A and B by verifying that P(AB) = P(A)P(B). Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

4 Conditional Probability If P(A) > 0, then the conditional probability of B given A is P(B A) = P(AB) P(A) Useful Interpretation: Think of P(B A) as the fraction of times B occurs among those in which A occurs Properties of Conditional Probabilities: 1 For any fixed A such that P(A) > 0, P( A) is a probability, i.e. it satisfies the rules of probability. 2 In general P(B A) P(A B) = P(B) Thus, another interpretation of independence is that knowing A does not change the probability of B. 3 If A and B are independent then P(B A) = P(AB) P(A) = P(A)P(B) P(A) Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

5 Law of Total Probability and Bayes Theorem Law of Total Probability Let A 1,..., A n be a partition of Ω, i.e. n i=1 A i = Ω (A 1,..., A n are jointly exhaustive events) A i Aj = for i j (A 1,..., A n are mutually exclusive events) P(A i ) > 0 Then for any event B P(B) = n P(B A i )P(A i ) i=1 Bayes Theorem Conditional probabilities can be inverted. That is, P(A B) = P(B A)P(A) P(B) Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

6 Random Variables We need the random variables to link sample spaces and events to data. A random variable is a mapping X : Ω R that assigns a real number X (ω) to each outcome ω Ω. This mapping induces probability on R from Ω as follows: Given a random variable X and a set A R, define and let X 1 (A) = {ω Ω : X (ω) A} P(X A) = P(X 1 (A)) = P({ω Ω : X (ω) A}) The cumulative distribution function (CDF) F X : R [0, 1] is defined by F X (x) = P(X x) CDF contains all the information about the random variable Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

7 Properties of CDFs Theorem A function F : R [0, 1] is a CDF for some random variable if and only if it satisfies the following three conditions: 1 F is non-decreasing: x 1 < x 2 F (x 1 ) F (x 2 ) 2 F is normalized: lim F (x) = 0 and lim F (x) = 1 x x + 3 F is right-continuous: lim F (y) = F (x) y x+0 Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

8 Discrete Random Variables X is discrete if it takes countable many values {x 1, x 2,...}. We define the probability mass function (PMF) for X by f X (x) = P(X = x) Relationships between CDF and PMF: The CDF of X is related to the PMF f X by F X (x) = P(X x) = f X (x i ) x i x The PMF f X is related to the CDF F X by f X (x) = F X (x) F X (x ) = F X (x) lim F (y) y x 0 Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

9 Continuous Random Variables A random variable is continuous if there exists a function f X such that f X (x) 0 for all x + f X (x)dx = 1, and For every a b P(a < X b) = b a f X (x)dx The function f X (x) is called the probability density function (PDF) Relationship between the CDF F X (x) and PDF f X (x): F X (x) = x f X (t)dt f X (x) = F X (x) Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

10 Transformation of Random Variables Suppose that X is a random variable with PDF f X and CDF F X. Let Y = r(x ) be a function of X. Q: How to compute the PDFand CDF of Y? 1 For each y, find the set A y = {x : r(x) y} 2 Find the CDF F Y (y) F Y (y) = P(Y y) = P(r(X ) y) = P(X A y ) = f X (x)dx A y 3 The PDF is then f Y (y) = F Y (y) Important Fact: When r is strictly monotonic, then r has an inverse s = r 1 and f Y (y) = f X (s(y)) ds(y) dy Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

11 Joint Distributions Discrete Case Given a pair of discrete random variables X and Y, their joint PMF is defined by Continuous Case f X,Y (x, y) = P(X = x, Y = y) A function f X,Y (x, y) is called the joint PDF of continuous random variables X and Y if + + fx,y (x, y) 0, f X,Y (x, y)dxdy = 1 For any set A R R P((X, Y ) A) = f X,Y (x, y)dxdy The joint CDF of X and Y is defined as F X,Y (x, y) = P(X x, Y y) A Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

12 Marginal Distributions Discrete Case If X and Y have joint PMF f X,Y, then the marginal PMF of X is f X (x) = P(X = x) = y P(X = x, Y = y) = y f X,Y (x, y) Similarly, the marginal PMF of Y is f Y (y) = P(Y = y) = x P(X = x, Y = y) = x f X,Y (x, y) Continuous Case If X and Y have joint PDF f X,Y, then the marginal PDFs of X and Y are f X (x) = f X,Y (x, y)dy and f Y (y) = f X,Y (x, y)dx Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

13 Independent Random Variables Two random variables X and Y are independent if, for every A and B Criterion of independence: Theorem P(X A, Y B) = P(X A)P(Y B) Let X and Y have joint PDF/PMF f X,Y. Then X and Y are independent if and only if f X,Y (x, y) = f X (x)f Y (y) Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

14 Conditional Distributions Discrete Case The conditional PMF: f X Y (x y) = P(X = x Y = y) = Continuous Case The conditional PDF is P(X = x, Y = y) P(Y = y) f X Y (x y) = f X,Y (x, y) f Y (y) = f X,Y (x, y) f Y (y) Then, P(X A Y = y) = f X Y (x y)dx A Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

15 Expectation and its Properties The expectation (or mean) of a random variable X is the average value of X. The expected value, or mean, or first moment of X is { µ X E[X ] = x xf X (x), xfx (x)dx, assuming that the sum (or integral) is well-defined. if X is discrete if X is continuous Let Y = r(x ), then E[Y ] = E[r(X )] = r(x)f X (x)dx If X 1,..., X n are random variables and a 1,..., a n are constants, then [ n ] n E a i X i = a i E[X i ] i=1 i=1 Let X 1,..., X n be independent random variables. Then, [ n ] n E X i = E[X i ] i=1 Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24 i=1

16 Variance and its Properties The variance measures the spread of a distribution. Let X be a random variance with mean µ X. The variance of X, denoted V[X ] or σx 2, is defined by { σx 2 V[X ] = E[(X µ X ) 2 ] = x (x µ X ) 2 f X (x), (x µx ) 2 f X (x)dx, if X is discrete if X is continuous The standard deviation is σ X = V[X ] Important Properties of V[X ]: V[X ] = E[X 2 ] µ 2 X If a and b are constants, then V[aX + b] = a 2 V[X ] If X 1,..., X n are independent and a 1,..., a n are constants, then [ n ] n V a i X i = ai 2 V[X i ] i=1 Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24 i=1

17 Covariance and Correlation If X and Y are random variables, then the covariance and correlation between X and Y measure how strong the linear relationship is between X and Y. Let X and Y be random variables with means µ X and µ Y and standard deviations σ X and σ Y. Define the covariance between X and Y by Cov(X, Y ) = E[(X µ X )(Y µ Y )] and the correlation by ρ(x, Y ) = Cov(X, Y ) σ X σ Y Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

18 Properties of Covariance and Correlation The covariance satisfies (useful in computations): Cov(X, Y ) = E[XY ] E[X ]E[Y ] The correlation satisfies: 1 ρ(x, Y ) 1 If Y = ax + b for some constants a and b, then { 1, if a > 0 ρ(x, Y ) = 1, if a < 0 If X and Y are independent, then Cov(X, Y ) = ρ(x, Y ) = 0. The converse is not true. For random variables X 1,..., X n [ n ] n V a i X i = ai 2 V[X i ] + 2 a i a j Cov(X i, X j ) i=1 i<j i=1 Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

19 Conditional Expectation and Conditional Variance The conditional expectation of X given Y = y is { E[X Y = y] = x xf X Y (x y), discrete case; xfx Y (x y)dx, continuous case. E[X ] is a number E[X Y = y] is a function of y E[X Y ] is the random variable whose value is E[X Y = y] when Y = y The Rule of Iterated Expectations EE[Y X ] = E[Y ] and EE[X Y ] = E[X ] The conditional variance of X given Y = y is V[X Y = y] = E[(X E[X Y = y]) 2 Y = y] V[X ] is a number V[X Y = y] is a function of y V[X Y ] is the random variable whose value is V[X Y = y] when Y = y For random variables X and Y V[X ] = EV[X Y ] + VE[X Y ] Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

20 Inequalities Markov inequality: If X is a non-negative random variable, then for any a > 0 P(X a) E[X ] a Chebyshev inequality: If X is a random variable with mean µ and variance σ 2, then for any a > 0 P( X µ a) σ2 a 2 Hoeffding inequality: Let X 1,..., X n Bernoulli(p), then for any ε > 0 P( X n p a) 2e 2na2 Cauchy-Schwarz inequality: If X and Y have finite variances, then Jensen Inequality: E[ XY ] E[X 2 ]E[Y 2 ] If g is convex, then E[g(X )] g(e[x ]) If g is concave, then E[g(X )] g(e[x ]) Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

21 Convergence of Random Variables There are two main types of convergence: convergence in probability and convergence in distribution. Let X 1, X 2,... be a sequence of random variables and let X be another random variable. Let F n denote the CDF of X n and let F denote the CDF of X. P X n converges to X in probability, written X n X, if for every ɛ > 0 lim P( X n X ɛ) = 0 n D X n converges to X in distribution, written X n X, if lim F n(x) = F (x) n for all x for which F is continuous. X n P D X implies that X n X Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

22 Law of Large Numbers and Central Limit Theorem The LLN says that the mean of a large sample is close to the mean of the distribution. The Law of Large Numbers Let X 1,..., X n be i.i.d. with mean µ and variance σ 2. Let X n = 1 n n i=1 X i. Then X n = 1 n n i=1 X i P µ as n The CLT says that X n has a distribution which is approximately Normal with mean µ and variance σ 2 /n. This is remarkable since nothing is assumed about the distribution of X i, except the existence of the mean and variance. The Central Limit Theorem Let X 1,..., X n be i.i.d. with mean µ and variance σ 2. Then Z n X n µ σ/ n D Z N (0, 1) as n Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

23 The Central Limit Theorem The central limit theorem tells us that Z n = X n µ σ/ n N (0, 1) However, in applications, we rarely know σ. We can estimate σ 2 from X 1,..., X n by sample variance Sn 2 = 1 n (X i X n ) 2 n 1 Question: If we replace σ with S n is the central limit theorem still true? Answer: Yes! Theorem i=1 Assume the same conditions as in the CLT. Then, X n µ S n / n D Z N (0, 1) as n Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24

24 Multivariate Central Limit Theorem Let X 1,..., X n be i.i.d. random vectors with mean µ and covariance matrix Σ: X 1i µ 1 E[X 1i ] X 2i µ 2 X i = µ =.. = E[X 2i ]. X ki µ k E[X ki ] V[X 1i ] Cov(X 1i, X 2i )... Cov(X 1i, X ki ) Cov(X 2i, X 1i ) V[X 2i ]... Cov(X 2i, X ki ) Σ = Cov(X ki, X 1i )... Cov(X ki, X k 1i ) V[X ki ] Let X n = (X 1n,..., X kn ) T. Then n(x n µ) D N (0, Σ) as n Konstantin Zuev (USC) Math 408, Lecture 11 February 11, / 24